The present invention concerns pattern-recognition machines, and specifically relates to the separation or segmentation of adjacent patterns.
In most electronic character-recognition machines, and in machines for recognizing other types of patterns, some means is required to separate adjacent patterns from each other, so that the recognition unit considers one and only one pattern at a time. Most optical character-recognition (OCR) machines, for example, contain a segmentation unit which detects gaps between characters, for gating an input video image to the recognition circuits. Some conventional OCR machines employ recognition-gated segmentation, in which the output of the recognition unit is gated as soon as this unit has successfully recognized some valid character.
Such conventional machines, however, tend to propagate segmentation errors. For example, a break in the middle of a "W" will cause at least two recognition errors, such as the gating of two outputs "VV". That is, most recognition machines find it difficult to recover from a single segmentation error.